In 1916, in Memphis, Tennessee, a grocer named Clarence Saunders opened a store called Piggly Wiggly and did something radical: he let customers fetch their own groceries. Until then, you handed your list to a clerk behind a counter, and the clerk walked the shelves for you. Saunders' self-service model won, and for the next century "service" increasingly meant doing it yourself: pumping your own gas, booking your own flights, and, eventually, logging into dozens of software applications to file your own expenses, update your own sales records and chase your own invoices. Now, quite suddenly, the clerk is coming back. This time the clerk is software, it never sleeps, and it can walk every shelf in every store at once. That is the agent layer, and it is beginning to reshape the industry known as SaaS.
The Kingdom of the Login Screen
Software as a Service, or SaaS, is the model most of us live inside without naming it. Instead of installing a program from a disc, you rent it by the month and use it through a browser. Salesforce, founded in 1999, made the idea famous with a campaign around the words "No Software" in the early 2000s, the cheeky promise that you would never install anything again. The model conquered business computing. Customer records, payroll, design, project management, accounting: almost every function of a modern company now lives in a rented web application. In August 2011, the investor Marc Andreessen summed up the era in a famous essay title: "Why Software Is Eating the World."
But SaaS carried a hidden assumption. Every application was designed for a human sitting in front of a screen. The dashboards, menus, buttons and forms were the product. Companies paid "per seat", meaning per human user, because the human was the one doing the work. The software was a beautifully organised workshop, but you still had to pick up the tools.
Enter the Agents
An AI agent is a system that does not just answer a question but carries out a task. Ask it to reconcile last month's invoices, and it opens the accounting system, compares the records against the bank statements, drafts emails about the discrepancies and asks you to approve them. It works across tools on your behalf, the way a capable assistant would, rather than waiting for you to click.
Two developments turned this from a demo into an industry. The first was capable language models that can plan, read documents and use software interfaces. The second was plumbing. In November 2024, Anthropic released the Model Context Protocol, an open standard for connecting AI systems to the tools and data they need, often described as a kind of USB port for agents. In 2025, Google introduced Agent2Agent, a protocol meant to let agents from different vendors talk to one another. Standards like these do for agents what common plugs did for electrical appliances: they make the ecosystem possible.
The incumbents noticed. Salesforce, the company that once promised "No Software", launched its Agentforce platform in 2024 and began pitching digital agents that work alongside human staff. Microsoft's chief executive, Satya Nadella, has argued publicly that the logic of business applications, which are at heart databases with business rules on top, is expected to move into the AI layer, with agents working across several databases at once.
From Seats to Outcomes
If the agent does the work, the logic of charging per human seat begins to wobble. Why pay for a hundred licences when ten people supervising a fleet of agents can get the same work done? The early answer from the market is a shift toward paying for outcomes rather than access. The customer-service company Intercom, for example, charges per resolution rather than per user.
This is the deeper meaning of "the agent layer". The value moves from the screen to the action. The interface, once the crown jewel of every software company, becomes optional: a place humans visit to supervise and approve, not to do. The database of record still matters enormously, since agents need somewhere trustworthy to read from and write to. But the battle for the customer's attention shifts to whoever controls the agent that orchestrates everything else.
The Risks of Letting the Clerk Behind the Counter
There is a reason we invented self-service: when you fetch the goods yourself, you can see exactly what you're getting. Agents reintroduce an intermediary, and intermediaries can make mistakes, be misled or be manipulated. Security researchers have already shown "prompt injection" attacks, in which hidden instructions planted in a web page or document trick an agent into doing something its user never intended. An agent with permission to send email, move money or change records is a far more attractive target than a chatbot that only talks.
Accountability blurs too. When an agent issues the wrong refund or deletes the wrong file, who is responsible: the user who delegated the task, the company that built the agent, or the vendor whose software it operated? And there is a quieter risk of dependence. If a single agent platform becomes the doorway to all of a company's tools, the lock-in that SaaS vendors once enjoyed could migrate to whoever owns that doorway.
What This Means for Publishers
The shift reaches beyond business software into how information itself is produced and consumed. More and more readers will meet our work through agents that search, summarise and cite on their behalf. For a public intelligence platform like Human-Level.AI, that changes the shape of the job. Articles, signals and archives can no longer be loose pages built only for human eyes. They have to become structured artifacts that humans review and approve, and that machines can then read, quote and verify reliably. Clear sourcing, stable addresses and visible editorial review stop being nice extras and become the interface.
Conclusion: The Return of the Clerk
The history of computing is a pendulum between doing things ourselves and having them done for us. SaaS put the tools in our hands; agents are taking them back, with the promise that we will spend less time clicking and more time deciding. Whether that future feels like liberation or loss of control depends on choices being made now: about permissions, accountability, open standards and who keeps the final say.
Clarence Saunders bet that people wanted to walk the aisles themselves. A century later, we are betting on the opposite. The question worth asking, before we hand over the shopping list, is simple: when the clerk returns, will we still know what is in the bag?

